AI platforms

Embedded Voice for AI Assistants

Add voice to an existing AI assistant product. Guava handles the call; your backend handles reasoning. One callback, your model in the loop.

Why this matters

Embed Guava in your AI stack. Your prompts, your data, our voice infrastructure. Sub-1s latency. WebRTC + SIP. Compliance you can hand to your enterprise buyer. This example shows you how to automate embedded voice for ai assistants calls end-to-end with Guava's SDK — no telephony plumbing, no prompt engineering, just Python.

How to set persona & task
Working with Fields and Say items
Handling callbacks
Running the agent
Step 01

Installation

Install from Guava's private PyPI index. A public package is coming soon — the install command will simplify to pip install guava.

terminal
# Step 1: Install Guava
pip install gridspace-guava --extra-index-url https://guava-pypi.gridspace.com

# Public PyPI package coming soon — the install command will simplify to:
# pip install guava

# Step 2: Set your credentials
export GUAVA_API_KEY="..."
export GUAVA_AGENT_NUMBER="..."
Step 02

How it's built

Every Guava agent is a Python class. Walk through the key sections below, then grab the complete file at the end.

01

Imports

Import the Guava SDK and any helpers you need. guava.CallController is the base class for every voice agent.

example.py
# Install: pip install gridspace-guava --extra-index-url https://guava-pypi.gridspace.com
import guava
import os
import requests
02

Agent setup

set_persona() defines how the agent presents itself. set_task() gives it its mission in plain English. The checklist drives the conversation — Guava works through it top-to-bottom, collecting Field values and speaking Say items as it goes.

example.py
class EmbeddedAssistantBot(guava.CallController):
    def __init__(self, assistant_id: str, user_id: str):
        super().__init__()
        self.assistant_id = assistant_id
        self.user_id = user_id
        self.thread_id = None
        self.set_persona(
            organization_name="Acme AI",
            agent_name="Atlas",
        )
        self.set_task(
            objective="Carry on an open-ended conversation, deferring reasoning to the customer's AI backend.",
            checklist=[
                "Greet the caller and ask how you can help today.",
                guava.OpenEnded(
                    description="Take any user utterance and call route_to_backend() to get the assistant's reply.",
                    handler=self.route_to_backend,
                ),
            ],
        )

    def route_to_backend(self, utterance: str) -> str:
        """Forward the user's words to the customer's AI backend and return its reply."""
        response = requests.post(
            "https://api.acme-ai.example/v1/threads/messages",
            json={
                "assistant_id": self.assistant_id,
                "user_id": self.user_id,
                "thread_id": self.thread_id,
                "message": utterance,
            },
            headers={"Authorization": f"Bearer {os.environ['ACME_API_KEY']}"},
            timeout=8,
        ).json()
        self.thread_id = response.get("thread_id")
        return response["reply"]

guava.listen(
    controller=EmbeddedAssistantBot,
    controller_args={"assistant_id": "asst_general", "user_id": "u_test_001"},
    agent_number=os.environ["GUAVA_AGENT_NUMBER"],
    api_key=os.environ["GUAVA_API_KEY"],
)

Platform performance

<1s

Response time

99.99%

Uptime SLA

13+

Industries served

Step 03

Full example

The complete file — copy it, save it as example.py, and run it.

example.py
# Install: pip install gridspace-guava --extra-index-url https://guava-pypi.gridspace.com
import guava
import os
import requests

class EmbeddedAssistantBot(guava.CallController):
    def __init__(self, assistant_id: str, user_id: str):
        super().__init__()
        self.assistant_id = assistant_id
        self.user_id = user_id
        self.thread_id = None
        self.set_persona(
            organization_name="Acme AI",
            agent_name="Atlas",
        )
        self.set_task(
            objective="Carry on an open-ended conversation, deferring reasoning to the customer's AI backend.",
            checklist=[
                "Greet the caller and ask how you can help today.",
                guava.OpenEnded(
                    description="Take any user utterance and call route_to_backend() to get the assistant's reply.",
                    handler=self.route_to_backend,
                ),
            ],
        )

    def route_to_backend(self, utterance: str) -> str:
        """Forward the user's words to the customer's AI backend and return its reply."""
        response = requests.post(
            "https://api.acme-ai.example/v1/threads/messages",
            json={
                "assistant_id": self.assistant_id,
                "user_id": self.user_id,
                "thread_id": self.thread_id,
                "message": utterance,
            },
            headers={"Authorization": f"Bearer {os.environ['ACME_API_KEY']}"},
            timeout=8,
        ).json()
        self.thread_id = response.get("thread_id")
        return response["reply"]

guava.listen(
    controller=EmbeddedAssistantBot,
    controller_args={"assistant_id": "asst_general", "user_id": "u_test_001"},
    agent_number=os.environ["GUAVA_AGENT_NUMBER"],
    api_key=os.environ["GUAVA_API_KEY"],
)
Step 04

Run it

Start the agent. It will connect to Guava's infrastructure and begin accepting calls on your assigned number.

terminal
python example.py
Get Started